Horizontal & vertical translations

crop and resize

blurring

brightness change

contrast change

mirroring

Dear Hunar,

there is a utility in OpenCV for data augmentation: opencv_createsamples. It is normally used for the creation of data for methods like Haar cascades, but can be employed for other things too.

opencv_createsamples has a lot of parameters, such as rotation angle and brightness and scale and skew factors, and also two different operating modes. I suggest you look into the tutorial below (even if it is a Haar tutorial) and read the documentation carefully: its default mode will insert background images into your samples. In medical images this is a thing that you do not want.

DOCUMENTATION (all those examples below describe a different version of opencv_createsamples) - https://docs.opencv.org/2.4/doc/user_guide/ug_traincascade.html - https://docs.opencv.org/3.0-beta/doc/user_guide/ug_traincascade.html Creating your own Haar Cascade OpenCV Python Tutorial: - https://pythonprogramming.net/haar-cascade-object-detection-python-opencv-tutorial/ OBSERVATION - MUST be copied to and run on the folder where the action occurs - Supposes there is a series of folders and files as per the documentation above

Good luck!

Aldo

Dear Ahmed, what is your main aim for enlarging your dataset? I believe technique may change in depend of your purpose

Dear Filippo Pesapane, I want to use the dataset for classification purpose using CNN, now I have 1000 samples with annotations, for that I want to make it double or triple by using augmentation techniques

Hi Hunar, for medical data standard approach would be doing slight rotations, translations, maybe jittering. You should absolutely take a look at elastic transformations (i.e. https://www.kaggle.com/bguberfain/elastic-transform-for-data-augmentation) that are very helpful specifically for medical CNNs and have been shown to improve training significantly.

thank you every one for your helpful answers.

We have utilized GAN to create synthetic data for augmentation purposes in the article that can be accessed at the:

Preprint Red blood cell image generation for data augmentation using ...

https://arxiv.org/abs/1901.06219

The "Related Work" section includes similar papers with focus on other domains including CT scan images.

you can try ImageJ (a java based software) having lots of plugins for data augmentation.